Model comparison
Codestral vs Qwen2.5-Coder-32B
Qwen2.5-Coder-32B is the stronger model overall, scoring 33.4 to 30.6 on the Noometry Index. Codestral costs 1.7× less per token, which makes it the better buy when Qwen2.5-Coder-32B's lead doesn't matter for your workload.
Last verified . 5 shared benchmarks.
Mistral AI
Alibaba (Qwen)Summary
- They share 5 benchmarks with published results for both. Codestral scores higher in 1 category and Qwen2.5-Coder-32B in 1 category; 2 gaps are clear of the uncertainty.
- The widest gap is in coding, where Codestral leads 27.3 to 22.6.
- The biggest single-benchmark swing is BigCodeBench Instruct: 41.8% for Codestral and 49% for Qwen2.5-Coder-32B.
- Codestral is cheaper at $0.30 / $0.90 per million input/output tokens, against $0.66 / $1 for Qwen2.5-Coder-32B.
- Codestral accepts more context: 256K tokens versus 33K.
- Qwen2.5-Coder-32B has downloadable open weights; the other is API-only.
Side by side
| Codestral | Qwen2.5-Coder-32B | |
|---|---|---|
| Provider | Mistral AI | Alibaba (Qwen) |
| Noometry Index | 30.6 | 33.4 |
| Released | 2024-05-29 | 2024-09-18 |
| Weights | Proprietary | Open |
| Context window | 256K | 33K |
| Max output | 8K | 29K |
| Input $ / M tokens | $0.30 | $0.66 |
| Output $ / M tokens | $0.90 | $1 |
| Results tracked | 7 | 31 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding Codestral leads
Codestral: 27.3 (#321), Qwen2.5-Coder-32B: 22.6 (#333)
| Benchmark | Codestral | Qwen2.5-Coder-32B |
|---|---|---|
| Aider Polyglot | 11.1% | 16.4% |
| BigCodeBench Instruct | 41.8% | 49% |
| BigCodeBench Complete | 52.5% | 58% |
| HumanEval+ | 73.8% | 87.2% |
| MBPP+ | 61.9% | 77% |
| SWE-bench Verified (bash only) | — | 9% |
| LiveBench Coding | — | 56.9% |
| LMArena Coding | — | 1276 |
| ALE-Bench | 137.78 | — |
Reasoning Qwen2.5-Coder-32B leads
Codestral: 19.8 (#251), Qwen2.5-Coder-32B: 21.2 (#225)
| Benchmark | Codestral | Qwen2.5-Coder-32B |
|---|---|---|
| Kagi LLM Benchmark | 32.5% | — |
| LiveBench Reasoning | — | 42.1% |
| LMArena Hard Prompts | — | 1251 |
| LiveBench Data Analysis | — | 49.9% |
| Epoch Capabilities Index | — | 119.49 |
| HellaSwag | — | 83% |
| LiveBench | — | 46.2% |
| WinoGrande | — | 80.8% |
Math Not comparable
Codestral: —, Qwen2.5-Coder-32B: 33.3 (#204)
| Benchmark | Codestral | Qwen2.5-Coder-32B |
|---|---|---|
| LiveBench Math | — | 46.6% |
| LMArena Math | — | 1251 |
| GSM8K | — | 93% |
Knowledge Not comparable
Codestral: —, Qwen2.5-Coder-32B: 33.4 (#203)
| Benchmark | Codestral | Qwen2.5-Coder-32B |
|---|---|---|
| LMArena Expert | — | 1221 |
| ARC (AI2) Challenge | — | 70.5% |
| MMLU | — | 79.1% |
Multilingual Not comparable
Codestral: —, Qwen2.5-Coder-32B: 37.8 (#235)
| Benchmark | Codestral | Qwen2.5-Coder-32B |
|---|---|---|
| LMArena Non-English | — | 1205 |
| LMArena Chinese | — | 1222 |
| LMArena Russian | — | 1228 |
Instruction Following Not comparable
Codestral: —, Qwen2.5-Coder-32B: 61.4 (#245)
| Benchmark | Codestral | Qwen2.5-Coder-32B |
|---|---|---|
| LiveBench Instruction Following | — | 58.7% |
| LMArena Instruction Following | — | 1223 |
Long Context Not comparable
Codestral: —, Qwen2.5-Coder-32B: 38.0 (#208)
| Benchmark | Codestral | Qwen2.5-Coder-32B |
|---|---|---|
| LMArena Longer Query | — | 1251 |
Writing & Preference Not comparable
Codestral: —, Qwen2.5-Coder-32B: 41.6 (#240)
| Benchmark | Codestral | Qwen2.5-Coder-32B |
|---|---|---|
| LMArena Text | — | 1230 |
| LMArena Creative Writing | — | 1174 |
| LMArena Multi-Turn | — | 1222 |
| LiveBench Language | — | 23.3% |
Frequently asked questions
Is Codestral better than Qwen2.5-Coder-32B?
Qwen2.5-Coder-32B is the stronger model overall, scoring 33.4 to 30.6 on the Noometry Index. Codestral costs 1.7× less per token, which makes it the better buy when Qwen2.5-Coder-32B's lead doesn't matter for your workload.
Which is cheaper, Codestral or Qwen2.5-Coder-32B?
Codestral is cheaper. It lists at $0.30 per million input tokens and $0.90 per million output tokens; Qwen2.5-Coder-32B lists at $0.66 and $1.
Is Codestral or Qwen2.5-Coder-32B better for coding?
Codestral scores higher on coding benchmarks: 27.3 versus 22.6 in the Noometry coding category.
Which has the bigger context window?
Codestral does, with 256K tokens against 33K.
How many benchmarks do Codestral and Qwen2.5-Coder-32B share?
5 benchmarks have published results for both models. Codestral has 7 scored results on Noometry and Qwen2.5-Coder-32B has 31.